Detalle de: Population and quantitative genetics for a cause

The Spanish Nobel Prize winner S. Ramón y Cajal (1852–1934) wrote that no pertinent scientific question has been exhausted by researchers, it is rather the researchers who become exhausted in answering the questions. This thought may well apply to the search of causative polymorphisms. At the annual meeting of the Society of Molecular Biology Evolution (http://www.smbe2008.com) recently held in Spain (Barcelona), it became clear to me that population and evolutionary genetics tend to employ more and more quantitative genetic concepts, whereas we quantitative geneticists are prone now to evolution and population genetics tools. Are animal breeders exhausted after so much quantitative genetics? Can population genetics help us to find the causative mutations? Animal breeding is, after all, but evolution at high speed – and Darwin knew it. I am convinced that strong heterosis awaits us at the crossroads between animal breeding (e.g. mixed-model technology) and population genetics (e.g. Tajima’s D and extended haplotype homozygosity concepts). A major contribution that animal breeding can lend to causal mutation discovery is a taste for statistical modelling of a phenotype. This should never be forgotten. A statistical model is not just a way to fit the data, it is a definition of a phenotype. Take just a naive example: two very distinct traits result when adjusting weight by age or weight by food consumption. Correcting by age gives growth and appetite, while correcting by food consumption is related to feed efficiency. When thousands of markers are analysed and the SNP P-values show a tight continuum, modelling can have a dramatic impact on discovering causative mutations. Yet in the whole of genome association studies there are case–control reports where little or no environmental variables are recorded. The phenotype can actually be a composite trait, partly caused by distinct environmental factors in each individual. Genetical genomics poses even harder problems for analysis and interpretation. Here optimum modelling becomes a non-polynomial time problem. But where is the line between population genetics and animal breeding? One key difference lies in the time scale. Population genetics counts in Ne (effective size) units, which means that populations, or breeds, evolve, recombine or undergo mutations that are scaled by Ne. The nucleotide variability found is the result of, on average, the last four Ne t generations, although with a very large variance. Thus the smaller the effective size, the shorter the time period that can be studied. It is kind of relativity phenomenon, the time slot is shorter the less people you have to recombine with. In contrast, animal breeders count real – actual – generations. Perhaps the most dramatic distinction between population and quantitative genetics relates to the selection coefficient s. In animal breeding textbooks, little time is spent in explaining the impact of Ne on response to selection, instead the selection coefficient s governs the increase in allele frequency of the favoured allele – full stop. It is taken for granted. However, the fate of a selected allele depends on the product Ne s, which means that random drift overcomes selection when the population is small. Symmetrically, for a positively selected allele to become fixed in a small population, its selective coefficient has to be just huge. An immediate consequence for livestock populations is that many alleles with moderate positive effects will be lost, whereas many neutral or deleterious alleles can become fixed. This of course may result in a lot of noise in genomewide association studies. Thus, it is hard to identify a mutation with tiny selection coefficient in a small population. Not only do we have the statistical problem of power, but genetic drift can also keep the favourable allele at low frequency. We can now start guessing why so few causative mutations have been found thus far in livestock populations, which are typically small. Another difference between population genetics and animal breeding theory is that population geneticists have traditionally focused on wild species, where controlled phenotypic information is not available. How can relevant genes be detected and studied without phenotypes? – an animal breeder wonders. The usual approach has been to compare the observed pattern of nucleotide variation and the value expected under the neutral model. To that end, a large number of selection tests have been developed, tests that have not been popular in animal breeding simply because enough molecular data have been lacking and because the QTL approach has been preferred – and useful enough because a direct link with phenotype is established. Now, as the QTL approach and ensuing association studies have also shown limitations, population genetics has emerged as a complementary tool for animal breeders. In animal breeding the main tool for detecting selection is to compare different breeds where phenotypic records are also available. Strong selection like that exerted in domestic species leaves a clear selective sweep signal on the haplotypes carrying a favourable mutation. A potential caveat of this approach is that the target of selection is not necessarily obvious. Thus, this signal in turn must be complemented with phenotypic information between and within populations. A second caveat is SNP selection ascertainment, i.e. the fact that SNPs are a biased sample of all extant SNPs in the population studied. This problem, however, will be overcome by next-generation sequencing technologies. A well-known hybrid product of animal breeding and population genetics theory is combining linkage and linkage disequilibrium (LDL) information. To model disequilibrium properly, one must resort to coalescence, and when modelling the phenotype properly, the mixed model theory is the best option. However, combined disequilibrium-linkage approaches have been replaced by large-scale genomewide association studies (GWAS). The reasons for the popularity of GWAS are the decreasing costs of genotyping and easy analyses in comparison with the LDL approach. In GWAS, direct association with phenotype is tested along the genome with a high marker density. The implicit assumption is that, with such a large number of markers, population pedigreeing does not add up to much information. This assumption has been surprisingly broadly accepted by the scientific community without too much criticism. However, given that a pedigree is but an amplification of a subset of all possible haplotypes in the founder individuals, controlled phenotyped individuals with known pedigree can help mutational discovery in at least two ways: (i) to improve accuracy in estimating the allelic effect, and (ii) to protect against unstable disequilibrium profiles and false-positive ratios due to stratification. Use of pedigree information allows us to increase power and decrease false discovery rate. In conclusion, animal breeding textbooks should include more population genetics; at least coalescence should be taught routinely. Let us face the challenge now of unifying disparate theories that were conceived to solve different problems but that, with current technologies, have converged to tackle the same question.